How PR impacts AI

If you work in PR right now, there's a good chance someone - a client, a stakeholder, a senior leader - has turned to you and asked some version of this question:

It's a fair question. AI chatbots like ChatGPT, Claude, and Gemini have become a routine part of how people find information, research products, and make decisions. And if that's where your target audiences are going to learn about the brands and businesses you represent, it's reasonable to want to know whether your media coverage is showing up there - and whether it's making a difference.

The good news is that it absolutely can. And the even better news is that you can measure it - without needing to be a data scientist or an AI expert.

This guide walks you through the AIMS framework: a step-by-step approach to measuring the AI impact of your PR campaigns. We'll explain the principles, walk through each step, and show you how to use it in your reporting.

AIMS stands for AI Impact Measurement Scale. It's a four-step framework that takes you from understanding your target audience all the way to scoring how likely your coverage is to be used by an AI chatbot.

The word AIMS is also a useful reminder of the approach: you start with the aims and objectives of your target audience - what do they want, what are they trying to find out, why are they going to AI chatbots in the first place? That thinking anchors everything else.

The first three steps - A, I, and M - work together to generate something we call the Campaign-Audience Prompts (CAP): the specific set of prompts that real people in your target audience are likely to ask an AI, in the context of your campaign. The fourth step - S - is where you measure how likely your coverage is to actually be used.

Let's walk through each one.

The starting point is always your audience - but not your general company audience. This is about the specific group of people you are targeting with this particular campaign.

This might already be well-defined in your campaign strategy. If so, great - you're ahead of the game. If not, it's worth taking the time to think it through properly, because everything else in the framework flows from here.

Start with the basics: demographics, behaviours, and context. Let's use an example that runs throughout this guide: imagine you're working on a campaign to promote Netflix's gaming service to casual gamers.

Your target audience might look something like this:

Once you know who they are, define what they actually need - specifically in relation to the campaign objective. These are the real-world desires, concerns, and questions that drive their behaviour.

For our Netflix gaming example:

From those needs, think about the specific tasks your audience will want to do to address them. These are the jobs-to-be-done that lead someone to reach for a chatbot:

Now that you know who your audience is, what they need, and what tasks they're trying to accomplish - you can think about what they will actually type into an AI chatbot.

This step is about generating a set of realistic, representative prompts. But here's something important to understand upfront: AI prompts are nothing like Google searches.

People don't type "Netflix gaming" or "streaming service games" into ChatGPT. They ask full questions, often with personal context, in natural conversational language. They provide background. They explain their situation. They ask follow-ups.

Taking our Netflix gaming audience and their tasks, here are the kinds of prompts they might actually type:

Notice how specific, personal, and context-rich these are. That's the nature of AI prompting - and it's important to capture that when you're building your list.

Each AI chatbot - ChatGPT, Claude, Gemini, Perplexity - is powered by its own AI model. And there are differences, not just between platforms, but within them: ChatGPT has several versions, and people use different ones.

More importantly: even when two different people type the same prompt, they may get different answers. That's because the AI model knows things about each user - from previous conversations, preferences they've shared, and contextual signals like their location or the time of year.

If you ask an AI chatbot to recommend a good restaurant in Canary Wharf, and it knows from your previous conversations that you're vegetarian, have young children, dislike chain restaurants, and are asking in the middle of summer - it's going to recommend something quite different from what it tells your colleague, who has an entirely different profile.

The Model step is a reminder that the prompt your audience types is not the full picture. The AI takes that prompt and layers in personal context before deciding how to respond - and which sources to draw on.

Once you've worked through Audience, Inputs, and Model, you have what we call the Campaign-Audience Prompts - or CAP for short.

The CAP is your working list of prompts: the specific questions that real people in your target audience are likely to ask an AI chatbot, in direct relation to your campaign objectives. It's where three things overlap:

It's worth noting what the CAP is not: it's not every possible prompt anyone could ever ask about your brand. Every brand has what we call a Brand Prompt Universe - the vast and ever-expanding range of questions people might ask AI about them. If you work in PR for Netflix, that includes everything from account settings and payment questions to share price queries and content scheduling.

Your CAP is a focused subset of that: only the prompts that are genuinely relevant to this campaign, for this audience. Everything outside that is interesting, but it's not what we're measuring.

One important distinction to build into your CAP: not all prompts mention the brand by name. Some people will search for your brand by name; others won't know it exists yet. Both types matter, and they map to different stages of the customer journey.

Both types should be in your CAP. Non-branded prompts reflect the early, discovery phase - where good coverage can introduce someone to your brand for the first time. Branded prompts reflect the consideration phase, where they're actively researching before making a decision.

It's worth being clear that success means something different depending on which type of prompt you're looking at.

For branded prompts, success is your coverage being used as a source in the AI's response - the article is cited or drawn on to answer a question about your brand or product.

For non-branded prompts, the AI uses your coverage as a source to answer the question - and because the question is specifically about the problem your campaign addresses, your brand gets mentioned or recommended in the response.

That distinction shapes how you write non-branded prompts. A useful test: ask yourself "what could someone in our target audience ask an AI, where the right answer would naturally include our brand?" Those are your non-branded prompts. If the answer to the question wouldn't logically mention your brand at all, the prompt probably belongs outside your CAP.

You now have your CAP - a focused, representative set of prompts your audience is likely to ask. The final step is the most important question: how likely is it that your coverage will actually be selected as a source when an AI responds to those prompts?

This is where we move from planning to measurement. And to understand it, we need to understand how AI models find and choose their sources.

Before anything else, can the AI even find your coverage?

Publishers decide whether they want AI bots to be able to read their content, and they communicate this via a file called robots.txt (you can see any site's version by adding /robots.txt to the domain). Some publishers welcome AI crawlers; others block them entirely.

This isn't always binary - some sites allow some AI bots but not others. And it's not a permanent judgement on the value of coverage: an article that is blocked by AI can still drive awareness, get shared on social media, and influence readers who then go on to use AI chatbots further down the funnel. But for the purpose of measuring AI impact specifically, blocked coverage scores zero.

Over time, knowing which publishers allow AI crawlers can genuinely inform your media strategy. Where you have a choice, publications that are open to AI bots will deliver more measurable AI impact.

Assuming your coverage is visible, the next question is: is it the right kind of source?

AI models are answer engines. When someone asks a question, the model is looking for the best possible source - and it has access to up to eight billion indexed web pages to choose from. Your media coverage is competing with all of them.

The principle is simple: a source that contains detailed, trusted, clearly written information - capable of specifically answering the question being asked - will win out over one that is vague, generic, or unreliable.

There are several content factors that determine how likely a piece of coverage is to be selected. Here are the most important ones:

It's a fair question. A list of eight criteria is useful context, but it doesn't tell you how to turn an article into a number.

At Releasd, we've built a scoring framework that automates this entirely. It uses AI to analyse the full text of each article and evaluate it against each of the criteria above, always in direct relation to a specific prompt from your CAP. Each factor is graded, the grades are weighted and combined, and the result is a single AI Impact Score per article per prompt. The underlying systems that drive this have taken a long time to develop and refine - but the output is fast, consistent, and comparable across campaigns.

If you want to try a simpler version yourself, you can. Here's a practical approach using any AI chatbot:

It's time-consuming at scale - which is exactly the problem Releasd solves - but for a handful of articles it's a useful way to get a feel for the framework in practice.

All of these factors - visibility and the specificity criteria above - can be combined into a single score for each piece of coverage, measured against each prompt in your CAP.

The score takes into account both whether the coverage can be found at all (visibility) and how strong a source it is when compared to what else is out there (specificity). A blocked article scores zero. A highly visible, original, authoritative, well-structured article that directly addresses the prompt in question scores highly.

Once you've run your coverage through the AIMS framework - whether manually or with a tool that does it automatically - you have a structured set of scores you can turn into meaningful insights for clients and stakeholders.

Here's the kind of analysis the framework makes possible:

Used well, this gives you a clear, credible story to tell: not just "we got X pieces of coverage," but "here's how that coverage is shaping the answers your audience is getting from AI right now."